Change Detection Analysis for Assessing the Vulnerability

PFG 2011 / 4, 0247–0260 ArticleStuttgart, August 2011

© 2011 E. Schweizerbart'sche Verlagsbuchhandlung, Stuttgart, Germany www.schweizerbart.deDOI: 10.1127/1432-8364/2011/0086 1432-8364/11/0086 $ 3.25

Change Detection Analysis for Assessing theVulnerability and Protective Effect of Beach Forests inCase of the Tsunami 2004 in Thailand

SebaStian Günthert, Marc Wieland & alexander SieGMund, Heidelberg

Keywords: Tsunami 2004, LULC classification, Post classification comparison, Statisticalmodelling

Summary: The influence of environmental fac-tors, especially of tropical coastal forests, on thedamage impact of a tsunami is still discussed con-troversially in literature. This paper focuses on theassessment of the vulnerability and protective ef-fect of beach forests in case of the Indian OceanTsunami event 2004 in Thailand, based on remotesensing and multivariate-statistical methods. Withthe use of a post classification comparison, we showthat the variability of the damage degree on differ-ent coastal vegetation types is highly spatial andspecies dependent. The combinatory evaluation ofresults from the change detection analysis and dataon coastal vegetation structure from a field surveyby using multiple regression analysis further provesthat beach forests can have a protective effectagainst tsunami waves, if they satisfy certain veg-etation structural conditions. In this context spe-cific vegetation parameters for the vulnerability aswell as the protective effect of beach forests are ac-quired, which mainly determine its vertical andhorizontal forest density. A concluding case studyfinally illustrates how a combination of empiricaltsunami hazard assessment and the presented workcould be utilized to support local coastal protectionin a targeted and efficient way.

Zusammenfassung: Change Detection-Analysezur Bewertung der Vulnerabilität und Schutzwir-kung von Strandwäldern im Falle des Tsunami2004 in Thailand. Der Einfluss biologischer Fakto-ren, vor allem tropischer Küstenwälder, auf dasSchadensausmaß eines Tsunami wird in der wis-senschaftlichen Literatur noch immer kontroversdiskutiert. Auf Grundlage von fernerkundlichenund multivariat-statistischen Methoden wird in dervorliegenden Arbeit die Vulnerabilität undSchutzwirkung von Strandwäldern im Falle desTsunamiereignisses in Thailand 2004 untersucht.Mittels eines Post Classification Comparison wirdaufgezeigt, dass die Variabilität des Schadensaus-maßes an verschiedenen Typen der tropischen Kü-stenvegetation stark von deren räumlicher Lagesowie der spezifischen Artzusammensetzung ab-hängt. Eine multiple Regressionsanalyse aufGrundlage der Change Detection-Analyse und vorOrt aufgenommenen Daten zur Vegetationsstruk-tur bestätigt ferner, dass Strandwälder eineSchutzwirkung gegenüber Tsunamis haben kön-nen, wenn sie bestimmte vegetationsstrukturelleBedingungen erfüllen. In diesem Zusammenhangwerden konkrete Vegetationsparameter sowohl fürdie Verwundbarkeit als auch für die Schutzwirkungvon Strandwäldern identifiziert, die hauptsächlichvon der vertikalen und horizontalen Walddichte ab-hängen. Ein abschließendes Fallbeispiel zeigtschließlich, wie die hier vorgestellte Arbeit in Kom-bination mit einer Tsunami-Gefahrenanalyse einenBeitrag zu einem gezielten und effizienten Küsten-schutz leisten kann.

248 Photogrammetrie • Fernerkundung • Geoinformation 4/2011

These maps allow a rapid identification ofzones with increased tsunami hazard and en-able furthermore a clear indication of the un-derlying factors, which determine the result-ing hazard level. They can therefore form thebasis for a more purposeful utilisation of thepresented results regarding coastal tsunamiprotection.

2 Research Methodology

The research at hand is based on a post classi-fication comparison of landuse and landcover(LULC) classifications before and after thetsunami 2004 (Fig. 1). After its implementa-tion, it is possible to derive detailed informa-tion about the spatial distribution and intensityof tsunami induced damages on differentcoastal forest types (Section 4).The results of the change detection analysis

and additional data from a terrestrial surveyare further used for multivariate statisticalmodelling with the objective to receive infor-mation about vegetation parameters, whichare of major importance for the vulnerabilityas well as for the protective function of beachforests (Section 5).On the basis of a case study, it is finally il-

lustrated, how the combination of both de-

1 Introduction and Objectives

The Indian Ocean Tsunami 2004 can be seenas a turning point in tsunami science. Neverbefore a tsunami disaster has been observed insuch detail. Waves of significantly differentheight and inundation distances were recordedalong the coastlines even in directly adjoiningbeach sections. These small scaled differenceswere mainly caused by variations of the near-shore-bathymetry in combination with hetero-geneous characteristics of the onshore topog-raphy (Wieland 2009). However, the influenceof certain factors on the impact of a tsunamiafter contact with land is still discussed con-troversially. This applies mainly for the possi-ble protective effect of environmental factors,such as beach forests. Their barrier effect,which is supposed to absorb and reflect thewave force as well as to decelerate the flowrate of a tsunami, depends on several complexfactors, which are characterized by co-de-pendence and interaction amongst them(Forbes & broadhead 2007). Especially dueto its specific vegetation structure concerningthe forest density and species composition aswell as its vulnerability (in the following de-fined as extent of destruction), not all coastalforests constitute an effective protection bar-rier.The paper at hand aims at two main aspects:

(1) detecting the overall impact of the tsunami2004 on different types of coastal vegetation(mangrove forests and various types of beachforests) in the study area and (2) providing acontribution for understanding the vulnerabil-ity and protective effect of especially beachforests in case of such waves.Moreover, the research is part of a coopera-

tive study focussing on the analysis of near-shore and on-shore geomorphological and en-vironmental factors, which can directly influ-ence the inundation width, damage extent anddeceleration of a tsunami – with the overallgoal of supporting local coastal protection intropical tsunami endangered regions. In thisconnection a tsunami hazard analysis (brieflypresented in Section 3) pointed out, which ge-omorphological factors are mainly influencingthe inundation width of a tsunami. The resultsare demonstrated in tsunami factorial mapsand hazard maps of the of the study area.

Fig. 1: Flowchart of the applied procedure foranalysing tsunami influencing factors and sup-porting local coastal protection in the studyarea.

Sebastian Günthert et al., Change Detection Analysis 249

the first waves reached the study area with itstouristic hotspots around Khao Lak andPhuket. Just minutes later also the coastal ar-eas of Krabi where hit by the incoming tsuna-mi. Due to refraction around the northern tipof Sumatra, the waves moved in an easterlydirection and hit the study area in a nearly per-pendicular angle. Overall velocity and energyof the approaching tsunami have been reducedby a relatively wide continental-shelf in frontof the Andamancoast. Variations of coastalconfiguration in combination with differingcharacteristics of the near-shore-bathymetrycaused waves of significantly different heightand characteristics along the coastline. Inmany regions of the study area, like in thenorthern and eastern part of Phuket or inPhang Nga bay, the tsunami occurred as ex-tensive flood-like inundation of low lyingcoastal areas, comparable to a rapidly increas-ing tidal gauge, reaching water heights from1–4m. In other parts, like in Patong andRawai Beach in the southern part of Phuket orin Khao Lak, wave heights of up to 12m abovesea level have been recorded. Therefore thedamages in the study area have been over-whelming in certain areas, but have not beendistributed homogeneously. Moreover, small-scaled differences in the degree of damagealong the coastline could be observed (Kel-letat & scheFFers 2006).Why are some areas more impacted by the

tsunami waves than neighbouring ones? Thefactors which according to literature have thegreatest influence on characteristics and im-pact of a tsunami can be divided into twogroups. The first group is well studied andconsists of morphological factors such as thewidth of the continental shelf, the near-shore-bathymetry (Murthy et al. 2007), offshore is-lands (bryant 2008), horizontal shape of thecoastline (Kong 2004) and the vertical profileof coastal topography (synolaKis & Kanoglu

1998). The second group, less studied so far, iscomposed of environmental factors like coralreefs (cochard et al. 2008, chatenoux &Peduzzi 2007), mangroves (TanaKa 2007) andbeach forests (cochard et al. 2008, Forbes &broadhead 2007).Based on GIS, remote sensing and multi-

variate-statistical methods in an integrated ap-proach, an empirical tsunami hazard analysis

scribed research activities, the empirical tsu-nami hazard analysis mentioned above and theassessment of the vulnerability and protectiveeffect of beach forests, can be used to supportlocal coastal protection in an efficient and costeffective way (Section 6).

3 Impact of the Tsunami 2004and its PotentialGeomorphological InfluencingFactors within the Study Area

The study area is located on the Malay Penin-sula along the west coast of Thailand, alsoknown as Andamancoast. It covers the coastalareas stretching from Ko Kho Khao in thenorth of Phang Nga province to the island ofPhuket and further south to Ko Phi Phi Don inKrabi province (Fig. 2).On December 26, 2004 at 7:58 h local time

an earthquake of magnitude 9.3 on the Richterscale occurred in the Sunda-trench in front ofthe coast of Sumatra and triggered the fataltsunami. Less than two hours after this event

Fig. 2: Location map of the study area, show-ing the Andamancoast in South-West-Thailand.(Data source: SRTM4)


ferent impacts along the coastline also takinginto account environmental factors like coralreefs.

4 Change Detection Analysis forIdentification of CoastalVegetation Damages

The assessing of coastal vegetation and in par-ticular tropical beach forests to possibly re-duce the inundation width and the tsunamis’hydraulic force onshore requires at first a com-prehensive analysis of the tsunami impactswithin the study area. For this purpose a

considered more than 30 mainly morphologi-cal influencing factors on the tsunami impact2004 on a regional scale (Wieland 2009). Theparameters, which could be identified as mostimportant on the impact of a tsunami, are cor-al reefs, wave exposition (including degree ofshelter from offshore islands), several factorsof coastal topography and near-shore-bathym-etry. Using the resulting regression-model tsu-nami factorial and hazard maps were createdfor this specific event (Fig. 3). The calculatedhazard maps show good overall correlationwith the real inundated areas, measured afterthe tsunami 2004. The factorial maps further-more give a possible explanation for the dif-

Fig. 3: Tsunami factorial (without nearshore bathymetry) and hazard map (calculated inundationareas) of Patong Beach and Kamala Beach in the south-western part of Phuket, Thailand.


30m to 15m spatial resolution by use of theHigh-Pass-Filter resolution merge (HPF-res-olution merge) was done. The algorithm pro-vides out-standing results concerning thepreservation of spectral information duringthe image fusion process (de Jong & Van der

Meer 2004, Mularz et al. 2000). This stepwas followed by the removal of clouds throughgenerating a binary mask. It was created byperforming a supervised classification of thesatellite imagery with two classes – “clouds”,“land/water” – and a subsequent GIS-basedenhancement (merging, buffering and elimi-nating of false classifications) of the classifica-tion results. Finally the NDVI as well asbrightness, wetness and greenness of the Tas-seled Cap Transformation (yarbrough et al.2005) and the first three bands of the PrincipalComponent Analysis (PCA) were calculatedin order to better separate different thematicclasses within the later supervised classifica-tion process.For a post classification refinement also a

digital geological map (scale 1:250.000) wasselected and an ASTER Digital ElevationModel (DEM) was implemented.

4.2 Multispectral LULC ClassificationBefore and After the Tsunami2004

The multispectral LULC classification of thepre- and post-tsunami satellite imagery, car-ried out within this study, consists of two mainsteps: a pixel based supervised classificationalgorithm, well suited for the classification ofsatellite images with medium spatial and widespectral resolution, and a knowledge basedpost classification refinement. Both techniqueswere performed with the software ERDASImagine 9.1. To support the selection of train-ing areas and the differentiation of classesduring the supervised classification process, aterrestrial survey in the study area for the as-sessing of ground control points was alreadyproceeded during a field trip in September andAugust 2008.The supervised classification was conduct-

ed by a hybrid classification technique. Theparallelepiped classification (also called boxclassifier) was chosen as first decision rule.

change detection analysis of medium resolu-tion satellite imagery (ASTER - AdvancedSpaceborne Thermal Emission and Reflect-ance Radiometer) on the basis of a post clas-sification comparison has been carried out inorder to get detailed information about thespatial distribution and intensity of tsunamiinduced damages, especially on beach forests.Even though being dependent on the quality ofthe two independently produced LULC clas-sifications (performed by a pixel-based clas-sification algorithm with following knowl-edge-based post classification refinement)used as input for the change detection, the postclassification comparison technique alreadyproofed capable of achieving very accurateoverall results (lu et al. 2004).Moreover, thismethod minimises impacts of atmospheric,sensor and environmental/seasonal differenc-es between multitemporal images, as thechange detection analysis is not directly basedon a comparison of the images’ spectral values(Mas 1999).

4.1 Base Data, Image Pre-Processingand Spectral Enhancement

For the presented analysis, the following level3 orthorectified ASTER images (P130/R54)with cloud cover < 5% and recording date asclose as possible to the tsunami event on26/12/2004 were used:

Pre tsunami: Two scenes from 31/01/2002,●two scenes from 07/03/2003Post tsunami: Two scenes from 31/12/2004,●two scenes from 08/02/2005

The ASTER sensor provides a wide spectralresolution with 14 different spectral bands(VNIR, band 1-4 / SWIR, band 4-9 / TIR,band 10-14) and a medium spatial resolutionbetween 15 and 90 m.To gain the maximum spatial and spectral

information content from the imagery and toreduce disturbing influences from clouds andcloud shadows (e. g., possible false classifica-tions due to its high spectral similarity withLULC classes like “infrastructure” and “baresoil” in the VNIR), different pre-processingand enhancement steps had to be performed.At first an image fusion of bands 4–9 from


the spatial distribution of previously identifiedthematic classes can be modified by includingfurther geodata like a DEM and geologicalmaps in the classification process and assign-ing specific thresholds for the different classesof interest. It underlies a hierarchical-deci-sion-tree-structure with (1) a hypothesis – theoutput class, (2) a rule – conditions for thevariables, which can be defined by the userand (3) variables – the thematic classificationand further geodata, from which characteris-tic values can be derived. For example, the fol-lowing rule was set up for beach forest: Beachforest can only occur near to the coast (< 500mdistance from the coastline), only on sandygrounds, under a slope of 8° and up to heightsof 10m above sea level (giesen et al. 2007),otherwise it has to be classified as tropicalrainforest.Finally, an accuracy assessment with 300

test points (selection method: equalized ran-dom with 20 samples per class) for every clas-sification was performed to statistically evalu-ate the quality of the LULC classifications

This according to Jensen 2005 very efficientclassification method assigns pixels with cor-responding spectral signatures to a thematicclass with a predefined signature interval. Allother signatures which could not thus be as-signed were then classified with the seconddecision rule, theminimum distance algorithm(cf. richards & Jia 2006).The computed thematic images comprised

32 different thematic classes in the pre-tsuna-mi- and 43 thematic classes in the post-tsuna-mi-classification. After subsequent post-pro-cessing by the use of recoding as well as filter-ing (fuzzy convolution filtering 3 x 3) for re-ducing salt and pepper effects, the final LULCclassifications consist of 15 thematic mainclasses (see Fig. 4).Due to close spectral similarities between

certain thematic classes (e. g., beach forest andtropical rainforest in shady mountain ridges)and resulting false classifications, a knowl-edge-based post classification refinement wascarried out afterwards. Using the so calledknowledge classifier in ERDAS Imagine 9.1

Tab. 1: Overall accuracy and omission/commission error for all thematic classes of the pre- andpost-tsunami LULC classification.

Thematic Classes Pre tsunami LULC classification Post tsunami LULC classification

*Pa **Ua ***Kc Pa Ua Kc

Beach forest 86,36 95 0,95 100 100 1

Open beach forest 100 100 1 100 90 0,89

Casuarina forest 100 90 0,89 100 100 1

Mangrove forest 90,91 100 1 95,24 100 1

Tropical rain forest 95,24 100 1 90,91 100 1

Sp. veg. trop. rain f. 94,74 90 0,89 80 80 0,79

Scrubland 100 95 0,95 95 95 0,95

Lawn 95 95 0,95 100 100 1

Agriculture 90,48 95 0,95 94,44 85 0,84

Plantations 95 95 0,95 75 90 0,89

Bare soil with sp. veg. 100 90 0,89 100 100 1

Bare soil 86,96 100 1 100 95 0,95

Rocks 93,75 75,00 0,74 100 90 0,89

Infrastructure 100 85 0.84 100 85 0,84

Overall accuracy: 93,67 % Overall accuracy: 94,00 %

* Producer’s accuracy ** User’s accuracy ***Kappa coefficient


curate acquisition of the locally different de-gree of damage on the coastal vegetation: theclassification of the “from-to” changes into adamage/degradation-scale with four differentlevels (see Tab. 2).It should generally be noted, that changes

between thematic classes are not necessarilycaused by the tsunami. Minor changes canalso be originated by clearings, constructionmeasures or natural effects of erosion betweenthe recording date of the pre-tsunami imagesand the tsunami event in December 2004.However, due to the short period of time be-tween the recording dates, these changes arenegligible compared with the impact of thetsunami.The damage extent of the coastal vegetation

and of all thematic classes detected by thechange detection analysis is presented exem-plarily in the following maps (Fig. 4), whichshow the small fishing village Thap Lamu inthe south of Khao Lak (see also Fig. 2).

4.4 Total Damage on CoastalVegetation in the Study Area

The four types of coastal vegetation in thestudy area have been affected by the tsunamiin a very different way. Fig. 5 shows the de-gree of damage on the different forest types intotal (ha) as well as in percent of the pre tsu-nami area.With almost 37.000 ha mangrove forest

presents by far the largest part of coastal veg-etation in the study area. Furthermore thisclass is significantly less affected (3.51% totaldegradation) by the tsunami than other class-

(richards & Jia 2006). Due to the implement-ed post classification refinement, high overallaccuracies for both classifications could beachieved (see Tab. 1).

4.3 Post Classification Comparisonof Pre- and Post-Tsunami LULCClassification

The computed LULC classification is suitablefor a first qualitative assessment of the tsuna-mi damages in the study area. However, for aquantitative analysis of the impacts, especiallywith regard to the extent of damage on the in-dividual beach forest classes, a change detec-tion analysis on the basis of a post classifica-tion comparison was needed. Within thischange detection technique, the classified im-ages are compared pixel wise by producing achange detection matrix (lillesand 2004).This generated layer provides the “from-to”information of every single thematic class,whereby its changes can be separately identi-fied and evaluated, depending on the subjectof investigation.

The changes between thematic classes inthe study area have only been detected in alimited spatial area: the theoretical tsunamihazard zone. It has been defined by the maxi-mum tsunami run-up height of 19m above sealevel and the maximum distance of tsunamiinundation of 3 km inland (chooWong et al.2007, cochard et al. 2008) within the studyarea.After implementation of the post classifica-

tion comparison, the generated informationhas been prepared in a way that allows an ac-

Tab. 2: Classification key for assigning the tsunami-induced LULC changes into different levels ofdamage.


or river estuaries. Thus they are rarely locatedin direct exposition to potential tsunamiwaves. As a result mangroves are less affect-ed, but also do, due to its preferred growinglocations, not seriously offer very much coast-al protection for settlements and infrastruc-ture in case of a tsunami.Beach forest is the second largest type of

coastal forest in the study area. This dense de-ciduous forest, consisting of different species

es. However, studies have shown that man-groves are usually strongly damaged if theyget in direct contact with tsunami waves (tan-aKa et al. 2006). Therefore the result of thechange detection analysis confirms the as-sumption of cochard et al. 2008 and chate-noux & Peduzzi 2007 that this type of coastalforest is, in contrast to the other presentedtypes, more likely to be found in areas whichare protected from the open sea, e. g., lagoons

Fig. 4: (1) Examples of the pre- and post-tsunami ASTER-imagery, (2) the computed LULC as wellas (3) the resulting tsunami damage maps.


vegetation parameters from 11 coastal testsites with a size of 2 ha.For the selection of the test sites during the

terrestrial survey, care was taken to ensure,that these areas were representative for thewhole beach section. After selecting a repre-sentative test area, following parameters havebeen assessed: Forest width, number of indi-viduals of the species, average height of thespecies, average height of the trees’ firstbranch, average diameter at breast height ofthe species, basal stem area of the species, av-erage distance between the individuals, cano-py cover, relative abundance and relative dom-inance of the species as well as type, heightand coverage of the undergrowth.Besides this, a second regression analysis

was carried out to get further informationabout the possible correlation between thevegetation structure of beach forests (8 testsites with a size of 2 ha) and damage on settle-ments (within an area of 4 ha directly behindthe test sites). This would be another indica-tion for the controversially discussed protec-tion function of beach forests in case of a tsu-nami (Section 5.2).

5.1 Multivariate-Statistical Operationsfor Detecting Influencing Factorson the Vulnerability of BeachForests

The regression analysis for evaluating thedamage on beach forests (dependent variable)within the 11 test sites implied four independ-ent variables shown in Tab. 3. By these para-meters, 94.8% of the damage variance couldbe explained (Adjusted R2: 0.984; F-test at asignificance level of 5%: 0.00; T-test at a sig-nificance level of 5%: ≤ 0.029; condition in-dex: ≤ 9.904). Its quantitative influence is ex-pressed in unstandardized coefficient (abso-lute influence on the dependent variable) andstandardized β-coefficient (relative influenceon the dependent variable).As expected, the local energy potential of

the tsunami, expressed by the run-up height,is mainly responsible (β-coefficient: 0.997) forthe extent of vegetation damage within the 11test sites. This run-up height is the result of thespatially different topographical and bathy-

of “barringtonia formation” (see cochard etal. 2008), has been damaged to 47.39%. Openbeach forest, a more sparsely vegetated foresthas been harmed to 66.21%. The smallest for-est type, the casuarina forest, only consists ofone species – Casuarina equitesifolia. With70.3% degradation in total, it was most af-fected by the tsunami waves. After tanaKa etal. (2006) this high proportion of damage canbe explained by the high vulnerability of espe-cially young casuarina populations (< 0.1mDiameter at Breast Height DBH). In situ in-vestigations confirmed this observation(günthert 2009).

5 Statistical Modelling of theVulnerability and ProtectiveEffect of Beach Forests

As already mentioned, the degree of damageon beach forests in the study area locally dif-fers a lot. To assess, whether these differentdamage pattern are solely attributed to the dif-ferent energy potential of the tsunami or alsodepend on the vegetation structure of the vari-ous forest sections, a linear multivariate re-gression analysis has been conducted (Section5.1). As data input three different sources weretaken: (1) On-site measured run-up heights ofthe tsunami 2004 (tsutJi et al. 2006), (2) dam-ages within the test sites as a result from thechange detection analysis and (3) more than 10

Fig. 5: Degree of damage on the differentcoastal forest types caused by the tsunami2004 in the study area.


rier (Forbes & broadhead 2007) whereby thewaves’ hydraulic force on the trees is attenu-ated and the damage on the forest is reducedby 3,38%.The “average distance between the single

species” (Fig. 6, green arrows) as an indicatorfor the horizontal forest density (β-coefficient:0.262) also influences the permeability andthus the damage on the forest (unstandardizedcoefficient: −2.591).

5.2 Multivariate-Statistical Operationsfor Detecting Influencing Factorson the Protective Effect of BeachForests

Several studies suggest the major role of beachforests as a protective barrier by absorbingand reflecting the waves’ energy as it passesthrough the forest (e. g., cochard et al. 2008,Forbes & broadhead 2007). By evaluating asecond multivariate statistic operation, thestudy at hand could confirm this point of viewabout the mitigation effect of beach forests bydetecting a statistically significant correlationbetween specific structural parameters within8 test sites and the damage on settlements be-hind them (3 test sites could not be integratedin the calculation because of the absence ofsettlements behind them). The regressionanalysis for evaluating the damage on settle-ments and infrastructure behind beach forests(dependent variable) contained two independ-ent variables (see Tab. 4), by which 93.3% of

metrical conditions on the shoreline. If itwould increase about one meter, the total dam-age of the coastal forest would rise about9.76% (see unstandardized coefficient, Tab. 3)The statistical model also verifies the sig-

nificant influence of structural parameters onthe vulnerability of beach forests. The aver-age height of the forest (Fig. 6, red arrows) isalmost equally important (β–coefficient:0.843) for its resilience than the run-up heightand can be considered as the main structuralrisk parameter. This coincides with studies ofcochard et al. (2008), who stated, that “themost important factor whether a tree survivesa tsunami wave (…) appears to be its size”. Itcan be assumed, that forest height is directlycorrelated with the forest age, tree diameterand in most cases with higher rooting depth ofthe trees. These factors directly enhance thebreaking strength of trunks and branches aswell as the risk of undercutting. But above acertain age, the resilience of a tree declinesbecause of the lowered elasticity of the treestems. This fact could not be considered in thepresented regression analysis.The variable “average height of the trees’

first branch” (Fig. 6, blue arrows) gives evi-dence about the vertical density within thestand. With increasing height of the firstbranch by 1m, the non-vegetated range fromthe ground to the canopy gets greater. As a re-sult the forest provides a more permeable bar-

Tab. 3: Variables and its calculated coefficientsof the regression analysis for evaluating thedamage on beach forests.

Variables UnstandardizedCoefficients

StandardizedCoefficients

Tsunamirunup height

9.76 0.997

Averageforest height

−2.784 −0.843

Averageheight of thetrees’ firstbranch

−3.383 −0.561

Averagedistance

between thespecies

−2.519 −0.262

Fig. 6: Most important vegetation parametersin a beach forest concerning its vulnerabilityand protective effect in case of a tsunami.


minalia catappa, Cocos nucifera) offeredmuch greater protection from the tsunami thanothers with only two species (Cocos nucifera,Casuarina equisetifolia), at a similar run-upheight (günthert 2009).

6 Potential AfforestationMeasures for Increasing theProtection from TsunamiWaves: The Case Study ofKamala Beach

The following case study shows how a combi-nation of empirical tsunami hazard assessment(see Section 3) and the presented work couldpossibly be applied to support local coastalprotection in a targeted and efficient way. Byusing tsunami factorial and hazard maps itwould be possible to rapidly identify zones ofincreased tsunami hazard together with itscauses. Within the identified zones the empiri-cally derived coefficients, presented in Sec-tions 5.1 and 5.2, could be utilized to calculatethe possible vulnerability and protective effectof beach forests by use of a regression equa-tion. If the mitigation effect of the specific for-est is considered to be low, the effectiveness ofpotential afforestation measures could also bestatistically modelled.The capacity of a beach forest to withstand

and buffer any tsunami impact strongly de-pends on the magnitude of the impact. In are-as, where the tsunami force was overwhelm-ing, the buffering role of vegetation was prob-ably fairly negligible (cochard et al. 2008).Therefore especially those areas with a lessseriously threat are preferred for afforestationmeasures.The test site of Kamala Beach (2 ha) (see

Fig. 2), an important tourist centre on the westcoast of Phuket island, is classified by the tsu-nami hazard map as an area of medium tsu-nami inundation hazard - the calculated inun-dation width amounts between 350–700m(Fig. 3). During the tsunami 2004 the test sitehas been hit by a wave of 4m height, wherebythe village of Kamala behind the test area gotdamaged by 66%. The coastal forest withinthe test site, which only consists of Casuarinaequisetifolia (see Tab. 5) was damaged by

the damage variance could be explained (Ad-justed R2: 0.933; F-test at a significance levelof 5%: 0.01; T-test at a significance level of5%: ≤ 0.008; condition index: ≤ 7.726).The “average height of the trees’ first

branch”, which gives information about thevertical density (as explained in Section 5.1),has the greatest influence on the damage ex-tent of the infrastructure behind the test sites(β-coefficient: 0.682). With an increase of theheight of the first branch, the damage on set-tlements would also increase about 6.1%. Thiscan be explained by the higher permeabilityand therefore reduced mitigation effects: astand with sparse undergrowth and trees withfew branches at lower levels (e. g., coconut for-est; old casuarina forest) has very low reflec-tion and energy absorption effects. As a resultit hardly reduces the inundation depth and hy-draulic force of the tsunami. The wave thusstreams almost unabated throughout the forestand unloads its whole energy on the infra-structure behind.The second variable within the regression

model, the “number of different species”(β-coefficient: −0.468), also influences the ver-tical density. With a larger number of speciesthe forest has a more heterogeneous structurewith different growth heights and habits. Sothe mitigation effect will increase by this vari-ation of the vertical stand structure and dam-age on settlements can severely be reduced(unstandardized coefficient: −10.132). Own in-vestigations on Phuket, Thailand show, that abeach forest with four different species (Cas-uarina equisetifolia, Pandanus tectorius, Ter-

Tab. 4: Variables and its calculated coefficientsof the regression analysis for evaluating thedamage on settlements and infrastructure be-hind beach forests.

Variables UnstandardizedCoefficients

StandardizedCoefficients

Averageheight of thetrees’ firstbranch

6.137 0.682

Number ofdifferentspecies

−10.132 −0.468


(damage reduction of 30.42%). At the sametime the damage on the beach forest wouldamount 51.81% (damage increase of 36.56%).Thus the protective effect of the beach for-

est could be increased significantly. Thewaves’ energy would be reduced by the im-proved absorbing and reflecting effect of theafforested vegetative barrier (resulting in agreater damage on the beach forest) and thesettlements behind are protected in a more ef-ficient way in case of a future tsunami.

7 Conclusions and Outlook

The described approach for a comprehensiveassessment of the vulnerability and protectiveeffect of tropical beach forests by using changedetection and terrestrial survey methods incombination with multivariate statistic model-ling proved to be very effective. However, dueto the limited number of test sites, the deter-mined statistical parameters can hardly beconsidered representative for other tropicalcoastal regions affected by the tsunami. Butits high statistical quality (adjusted R2: > 93%)leads still to the assumption, that this studyprovides valid evidences for the significanceof specific vegetation structures, which deter-mine whether a tropical beach forest providesa protective effect in case of a tsunami or not.Attention should also be paid for the forest’svulnerability. The buffering effect of vegeta-tion barriers is always accompanied by itspartly or full destruction and can in the lattercase lead to an even greater threat for settle-ments by increased debris flow.Further investigations should now be done

to validate and enhance the statistical modelsfor the vulnerability and protective effect of

15.25%. (all damage parameters are based onthe results of the change detection analysis,but were also similarly calculated by the re-gression equation).Overall, this very old forest has a low verti-

cal and horizontal density and therefore didnot offer sufficient protection against the tsu-nami waves. To gain a better protective effectof this vegetation barrier, afforestation meas-ures have to be conducted, but in a way thatallows a state of balance between vulnerabili-ty and protection. If the forest would be affor-ested too dense (calculated theoretical damageon settlements of 0%), a large wave may com-pletely level the forest and pass over unmiti-gated. This effect is usually accompanied byits total destruction (damage on forest of100%), resulting in an additional hazard forthe coastal village from the debris laden water(Forbes & broadhead 2007).So on the one hand the desired afforestation

project needs to improve the green belts’ verti-cal density by replacing a part of the casuarinaspecies by smaller trees with a proven robust-ness against tsunami waves – like pandanustectorius, a species with supported aerialroots. On the other hand, additionally plantedtrees of hibiscus tiliaceus which are cultivatedfor tsunami protection in Japan, directly en-hance the horizontal density of the forest aswell as the vertical stand structure.Tab. 6 shows the vegetation structure of the

test site after potential afforestation measures(the several parameters of the new specieshave been recorded in a similar manner onother test sites, the same applies for the newspecies composition and its species propor-tion).The statistical modelling of the vulnerabil-

ity and protective effect of the afforested testsite led to the following results: The damageon settlements would now amount 35.58%

Tab. 5: Vegetation structure of the test site“Kamala Beach”.

Species

Num

bero

fin

divi

dual

s

Øh

eigh

tin

m

Øh

eigh

tof

firstbranch

Ø-distance

betw

een

indi

vidu

als

Casuarina e. 347 39 4,5 3,6

Tab. 6: Vegetation structure of the test site“Kamala Beach” after afforestation measures.

Species

Num

bero

fin

divi

dual

s

Øh

eigh

tin

m

Øh

eigh

tof

firstbranch

Ø-distance

betw

een

indi

vidu

als

Casuarina e. 209 39 4,5

2,5Pandanus t. 138 8,6 1,85

Hibiscus t. 150 12 2


raphische Rundschau International Edition 2 (1):4–9.

Kong, l., 2004: Oceanography Special Report. –ioc3.unesco.org/itic/files/worldbook_tsunami.pdf (11.11.2008).

lillesand, t., KieFer, r. & chiPMan, J., 2004: Re-mote Sensing and Image Interpretation. – 5thed., Wiley, Hoboken, NJ, USA.

lu, d., Mausel, P., brondízio, e. & Moran, e,2004: Change detection techniques. – Interna-tional Journal of Remote Sensing 25 (12): 2365–2407.

Mas, J.-F., 1999: Monitoring land-cover changes: acomparison of change detection techniques. –International Journal of Remote Sensing 20(1):139–152.

Mularz, s., drzeWiecKi, W. & PiroWsKi, t., 2000:Merging Landsat TM Images and airborne pho-tographs for monitoring of open-cast mine area.– International Archives of Photogrammetryand Remote Sensing 33 (B7): 920–927.

Murthy, K.s.r., Murty, g., rao, K & subrahMan-yaM, V., 2007: Impact of coastal morphology,structure and seismicity on the tsunami surge. –The Indian Ocean Tsunami. London, Taylor andFrancis.

richards, J.a. & Jia, x., 2006: Remote SensingDigital Image Analysis. – 4th ed., Springer, Ber-lin.

synolaKis, c.e. & Kanoglu, u., 1998: Long waverunup on piecewise linear topographics. – Jour-nal of Fluid Mechanics 374: 1–28.

tanaKa, n., sasaKi, y., MoWJood, M., Jinadasa, K.& hoMchuen, s., 2006: Coastal vegetation struc-tures and their functions in tsunami protection:experience of the recent Indian Ocean tsunami.– Landscape and Ecological Engineering 3 (1):33–45.

tsuJi, y., naMegaya, y., MatsuMoto, h., iWasaKi,s.-i., Kanbua, W., sriWichai, M. & Mee-suK, V.,2006: The 2004 Indian tsunami in Thailand:Surveyed runup heights and tide gauge records.– Earth, Planets and Space 58: 223–232.

Wieland, M., 2009: Einsatz von GIS undFernerkundung in der Katastrophenvorsorge:Analyse und Bewertung tsunamigefährdeterKüstenbereiche entlang der Westküste Thailan-ds. – Unpublished Diploma Thesis. University ofHeidelberg.

yarbrough, l., easson, g. & KuszMaul, J., 2005:Using At-Sensor Radiance and Reflectance Tas-seled Cap Transforms Applied to Change Detec-tion for the ASTER Sensor. – IEEE Third Inter-national Workshop on the Analysis of Multi-temporal Remote Sensing Images.

beach forests from the study at hand. In thiscontext, it would also be interesting to exploreits transferability to other tsunami endangeredregions like Chile and/or Japan.

References

bryant, e., 2008: Tsunami: The Underrated Haz-ard. – 2nd ed., Cambridge Univ. Press., Cam-bridge.

chatenoux, b. & Peduzzi, P., 2007: Impacts fromthe 2004 Indian Ocean Tsunami: analysing thepotential protecting role of environmental fea-tures. – Natural Hazards 40: 289–304.

chooWong, M., MuraKoshi, n., hisada, K., charu-siri, P., daorerK, V., charoentitirat, t., chuta-KositKanon, V., JanKaeW, K. & KanJanaPayont,P., 2007: Erosion and Deposition by the 2004 In-dian Ocean Tsunami in Phuket and Phang-ngaProvinces, Thailand. – Journal of Coastal Re-search 23 (5): 1270–1276.

cochard, r., ranaMuKhaarachchi, s., shiVaKoti,g., shiPin, o., edWards, P. & seeland, K., 2008:The 2004 tsunami in Aceh and Southern Thai-land: A review on coastal ecosystems, wave haz-ards and vulnerability. – Perspectives in PlantEcology, Evolution and Systematics 10: 3–40.

de Jong, s.M. & Van der Meer, F.d., 2004: RemoteSensing Image Analysis. Including the SpatialDomain. – Springer Netherlands, Dordrecht,NE.

Forbes, K. & broadhead, J., 2007: The role ofcoastal forests in the mitigation of tsunami im-pacts. – RAP Publications 2007 (1), FAO Re-gional Office for Asia and the Pacific, Bangkok,Thailand.

giesen, W., WulFrat, s., zieren, M. & schollen,l., 2007: Mangrove Guidebook for SoutheastAsia. – RAP Publication 2006/7. FAO RegionalOffice for Asia and the Pacific, Bangkok, Thai-land.

günthert, s., 2009: Untersuchungen zur Vulnera-bilität und Schutzfunktion von Strandwäldernbei Tsunami-Ereignissen: Eine GIS- undfernerkundungsgestützte Analyse am Beispieldes Tsunami vom 26.12.2004 in Thailand. – Un-published Diploma Thesis. University of Heidel-berg.

Jensen, J.r., 2005: Introductory Digital ImageProcessing: A Remote Sensing Perspective. –3rd ed., Pearson Prentice Hall, Upper SaddleRiver, NJ, USA.

Kelletat, d., scheFFers, s. & scheFFers, a., 2006:Learning from the Southeast-Asian Tsunami:Examples from Thailand s West Coast. – Geog-


Addresses of the Authors:

Dipl.-Geogr. sebastian günthert, Prof. Dr. alex-ander siegMund, University of Education & Uni-versity Heidelberg, Department of Geography.D-69115 Heidelberg, Phone: +49-6221-477-795,-771, e-mail: [email protected], [email protected]

Dipl.-Geogr. Marc Wieland, GFZ German Re-search Centre for Geosciences, Department 2 Phys-ics of the Earth, Section 2.1 Earthquake Risk andEarly Warning. D-14473 Potsdam, Phone: +49-331-288-1283, e-mail: [email protected]

Manuskript eingereicht: Februar 2011Angenommen: April 2011

Documents

Change Detection Analysis for Assessing the Vulnerability